Unverified50% confidenceFactExact time
In Soma v3, the fitness calculation for blocking rules was inverted: the more bad commits a gate blocked, the worse its score looked.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
Unverified底层模型更换后，即便微小的输出分布变化也可能导致此前校准的安全护栏过度拦截或防护失效，需要重新标注和调试68% similarUnverified循环的好坏取决于验证机制和完成标准，主观评分标准会导致迭代难以逼近参考目标67% similarUnverifiedThe success rate of relying on large language model judgment to reject malicious prompt injections is unstable, and scope restrictions serve as a fallback mechanism when model judgment fails.66% similarVerifiedThe fundamental reason prompt injection is difficult to defend against is that LLMs cannot reliably distinguish between data and instructions.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1062385API
curl https://kongchang.com/api/v1/knowledge/claims/1062385MCP
get_claim(id=1062385)